Document 10524210

advertisement
127 (2002)
MATHEMATICA BOHEMICA
No. 4, 581–589
THE PICONE IDENTITY FOR A CLASS OF PARTIAL
DIFFERENTIAL EQUATIONS
, Brno
(Received February 16, 2001)
Abstract. The Picone-type identity for the half-linear second order partial differential
equation
n ∂
∂u
Φ
+ c(x)Φ(u) = 0, Φ(u) := |u|p−2 u, p > 1,
∂xi
∂xi i=1
is established and some applications of this identity are suggested.
Keywords: Picone’s identity, half-linear PDE, p-Laplacian, variational technique
MSC 2000 : 35J20
1. Introduction
The classical Picone identity established by Picone [24] concerns a pair of second
order Sturm-Liouville operators
(1)
l[x] := (r(t)x0 )0 + c(t)x,
0 0
L[y] := (R(t)y ) + C(t)y,
r(t) > 0,
R(t) > 0
and reads as follows. Let x, y be differentiable functions such that rx0 , Ry 0 are also
differentiable and y(t) 6= 0 on an interval I ⊂ . Then in this interval
x 2 x
d x
(yrx0 − xRy 0 ) = (r − R)x02 + (C − c)x2 + R x0 − y 0 + {yl[x] − xL[y]} .
dt y
y
y
Research supported by the Grant A1019902 of the Grant Agency of the Czech Academy
of Sciences (Prague).
581
Since 1910, when the original paper of Picone was published, this identity has been
extended to various equations (not only to ODE’s but also to PDE’s and to difference
equations, see [1], [3], [12], [13], [14], [16], [17], [23], [27] and the references given
therein), and it has turned out to be a very useful tool in the oscillation theory of
these equations.
In our paper we follow this idea and establish a Picone-type identity for the second
order partial differential equation
n
X
∂u
∂
+ c(x)Φ(u) = 0,
(2)
Φ
∂xi
∂xi
i=1
where Φ(u) := |u|p−2 u, p > 1, x = (x1 , . . . , xn ) ∈ n and c : n → is a Hölder
continuous function (Theorem 1). We also present some basic consequences of this
identity (Theorems 2–4).
∂u
∂u
If p = 2, i.e. Φ( ∂x
) = ∂x
, then (2) reduces to the classical Laplace equation
i
i
(3)
∆u + c(x)u = 0.
There also exists a voluminous literature dealing with the PDE’s with the so-called
p-Laplacian
(4)
div(k∇ukp−2
∇u) + c(x)Φ(u) = 0,
2
∂u
∂u
where ∇ is the usual nabla operator ∇u = ( ∂x
, . . . , ∂x
) and kxk2 =
1
n
n
P
i=1
x2i
12
is the Euclidean norm in n , see e.g. [7], [11] and the references given therein. In
these papers and monographs, it was shown that solutions of (4) have many of the
properties typical for the equation with the classical Laplacian (3). Let us mention
at least the papers [1], [2], [10], [12], [14] dealing with Picone’s identity for (4). This
identity is used in [1], [2] to study the Dirichlet eigenvalue problem associated with
(4) and the papers [10], [12], [14] deal with the application of Picone’s identity in
oscillation theory and show that oscillatory properties of (4) are very similar to those
of (3).
PDE of the form (2) has been investigated in a series of papers of G. Bognár [4],
[5], [6], and basic properties of the eigenvalue problem associated with (2) have been
established. Our investigation can be regarded as a continuation of this research and
also as an attempt to extend the results concerning (3) and (4) to (2). In particular,
using the Picone identity established in this paper we show that the basic facts of
the classical Sturmian theory (like Sturm comparison and separation theorems) do
extend to (2).
582
2. The Picone identity
Let Ω ⊂ n be a bounded domain and let u ∈ W01,p (Ω). Then multiplying the
left-hand side of (2) by u and integrating the obtained formula over Ω (using the
Gauss theorem) we see that (2) is closely related to the p-degree functional
(5)
Fp (u; Ω) :=
where kxkp =
Z X
Z
n ∂u p
k∇ukpp − c(x)|u|p dx,
− c(x)|u(x)|p dx =
∂xi
Ω
Ω
i=1
n
P
|xi |p
i=1
p1
denotes the p-norm in n
. Another important object
associated with (2) is a Riccati-type equation which we obtain as follows. Let u be
a solution of (2) which is nonzero in Ω and denote
v :=
∂u
∂u
,...,Φ
,
Φ
∂x1
∂xn
w :=
v
.
Φ(u)
Then, using the fact that (2) can be written in the form div v = −c(x)Φ(u), we
have
div w =
1
Φ2 (u)
{Φ(u) div v − Φ0 (u)h∇u, vi}
n
|u|p−2 X ∂u p
|u|2p−2 i=1 ∂xi
n X
∂u/∂xi q
= −c(x) − (p − 1)
Φ
u
i=1
= −c(x) − (p − 1)
= −c(x) − (p − 1)kwkqq ,
where h·, ·i denotes the usual scalar product in n , q :=
n
1
P
of p and kxkq =
|xi |q q denotes the q-norm in p
p−1
n
is the conjugate exponent
. Consequently, the vector
i=1
variable w satisfies the Riccati-type equation
(6)
div w + c(x) + (p − 1)kwkqq = 0.
Of course, if p = q = 2 and n = 1, then the last equation reduces to the classical
Riccati equation w 0 + c(t) + w2 = 0 associated with (1) (where r(t) ≡ 1) via the
substitution w = u0 /u. Recall also that if we apply the same procedure as above to
equation (4) and the vector variable z := ∇u
u , we get the functional
(7)
Fep (u; Ω) :=
Z
Ω
{k∇ukp2 − c(x)|u|p } dx
583
and the Riccati-type equation
div z + c(x) + (p − 1)kwkq2 = 0.
(8)
Now we are in a position to formulate a Picone-type identity for (2), more precisely,
for the p-degree functional (5) and the Riccati type equation (6) associated with (2).
Before formulating it, observe that if R ≡ r, C ≡ c in (1), y is a solution of l(y) = 0
0
and w := ryy , then Picone’s identity for (1) mentioned in the first section can be
written in the form
r(t)x0 2 − c(t)x2 = [x2 w(t)]0 +
1
[r(t)x0 − w(t)x]2 ,
r(t)
and integrating this identity over I = [a, b] we get the integral form of this identity
(9)
F(x; a, b) :=
Z
b
[r(t)x0 2 − c(t)x2 ] dt
a
b
= x w(t)a +
2
Z
b
r−1 (t)[r(t)x0 − w(t)x]2 dt.
a
Theorem 1. Let w be a solution of (6) which is defined in Ω (closure of Ω) and
u ∈ W 1,p (Ω). Then
(10)
Z
|u(x)|p w(x) dS
Z k∇u(x)kpp
kw(x)kqq |Φ(u(x))|q
+p
dx.
− h∇u(x), Φ(u(x))w(x)i +
p
q
Ω
Fp (u; Ω) =
∂Ω
Moreover, the last integral in this formula is always nonnegative and for a function
u 6≡ 0 it equals zero only if u =
6 0 in Ω and
(11)
w=
1
∂u
∂u
,...,Φ
.
Φ
Φ(u)
∂x1
∂xn
. By a direct computation, using (6), we have
div (w|u|p ) = |u|p div w + pΦ(u)h∇u, wi
= |u|p −c(x) − (p − 1)kwkqq + k∇ukpp − k∇ukpp + ph∇u, Φ(u)wi
k∇ukpp
kwkqq |Φ(u)|q
= k∇ukpp − c(x)|u|p − p
− h∇u, Φ(u)wi +
.
p
q
Integrating the last formula over Ω and using the Gauss theorem we get (10).
584
To prove nonnegativity of the second integral in (10), consider the function f (x) =
Its conjugate function
kxkp
p
p .
f ∗ (y) := sup {hx, yi − f (x)}
(12)
x∈
n
kykq
is f ∗ (y) = q q as can be verified by a direct computation. The supremum in (12)
is attained for
(13)
x = (x1 , . . . , xn ) = (Φq (y1 ), . . . , Φq (yn )) ,
Φq (s) := |s|q−2 s
being the inverse function of Φ(s) = |s|p−2 s. The classical Fenchel inequality (see
[25]) now implies that
kykqq
kxkpp
− hx, yi +
>0
p
q
for every x, y ∈ n with equality if and only if x and y are related by (13). Substituting there x = ∇u, y = Φ(u)w we see that the (continuous and nonnegative)
integrand in the second integral of (10) equals zero if and only if u and w are related
by (11).
Substituting p = 2 in the previous theorem, we get Picone’s identity as formulated
e.g. in [27] and one can immediately see its similarity with (9).
Using Picone’s identity (10) we can now prove the following relationship between
the positivity of Fp (over W01,p (Ω)) and the existence of a nonzero solution (in Ω) of
(2).
Corollary 1. If there exists a solution u = u(x) of (2) such that u(x) 6= 0 for
x ∈ Ω then Fp (y; Ω) > 0 for every 0 6≡ y ∈ W01,p (Ω).
. The proof immediately follows from Theorem 1 by setting
∂u
∂u
1
,...,Φ
w=
Φ
Φ(u)
∂x1
∂xn
and using the fact that w and ∇y cannot be proportional since u(x) > 0 in Ω and
y|∂Ω = 0.
585
3. Applications
Following the usual terminology for linear equation (3) (see e.g. [21], [22]), a
bounded domain Ω ⊂ n is said to be the nodal domain of a nontrivial solution
u of (2) if u|∂Ω = 0, and the curve Γ = {x ∈ Ω : u(x) = 0} is called the nodal
contour of u in Ω. Corollary 1 immediately implies the following Sturmian-type
statement.
Theorem 2. Suppose that u is a nontrivial solution of (2) with the nodal domain
Ω ⊂ n and c̃ : n → is a Hölder continuous function such that c̃(x) > c(x) in Ω.
Then every solution of the (majorant) equation
n
X
∂u
∂
+ c̃(x)Φ(u) = 0
Φ
∂xi
∂xi
i=1
(14)
has a nodal contour in Ω. In particular, if c̃(x) = c(x), then every solution of (2) has
a nodal contour in Ω.
.
If ũ is a solution of (14) without a nodal contour in Ω, then by Corol-
lary 1
Z
Ω
k∇ykpp − c̃(x)|y|p dx > 0
for every nontrivial y ∈ W01,p (Ω) and in view of the inequality c̃(x) > c(x) in Ω,
we have Fp (y; Ω) > 0 for every nontrivial y ∈ W01,p (Ω). On the other hand, the
solution u of (2) satisfies u ∈ W01,p (Ω) and by the Gauss theorem Fp (u; Ω) = 0, a
contradiction.
Corollary 1 also suggests another application in the oscillation theory of (2),
namely to compare this equation with a certain half-linear ODE. To prove that
(2) possesses no positive solution in a connected (bounded or unbounded) domain
Ω ⊂ n , it suffices to find a nontrivial function y ∈ W01,p (Ω) for which Fp (y; Ω) 6 0.
It is natural to look for such a function in the radial form y(x) = ỹ(kxk), where k · k
is a norm in n and ỹ : → . This idea has been used in [10], [14] for PDE with
p-Laplacian (4) with the norm k · kp , for our equation (2) it is convenient to take the
q-norm k · kq .
Theorem 3. Let
C(r) :=
586
Z
c(x) dS,
kxkq =r
q=
p
.
p−1
If the half-linear second order ODE
(15)
r
n−1
0
1
C(r)Φ(z) = 0,
Φ(z ) +
ωn (q)
0
0
d
:=
, ωn (q) :=
dr
Z
dS
kxkq =1
is oscillatory (i.e., every nontrivial solution of (15) has arbitrarily large zeros), then
(2) possesses no positive solution in the exterior domain ΩR = {x ∈ n : kxkq > R}
for every R > 0.
. If (15) is oscillatory, there exists an increasing sequence rn → ∞ and
a nontrivial solution z = z(r) of (15) for which z(rn ) = 0. Let R > 0 be arbitrary.
Then rn > R for n ∈ sufficiently large and for some of these n’s define
y(x) =
Then, if Ωrn ,rn+1 := {x ∈ Fp (y; Ωrn ,rn+1 ) =
=
Z
Z
Ωrn ,rn+1
rn+1
rn
= ωn (q)
n
Z
Z
(
z(kxkq ),
rn 6 kxkq 6 rn+1 ,
0
elsewhere.
| : rn 6 kxkq 6 rn+1 }, we have
k∇ykpp − c(x)|y|p dx
kxkq =r
rn+1 [k∇ykpp
− c(x)|y| ] dS dr
C(r)
|z(r)|p dr
ωn (q)
Z rn+1 C(r)
Φ(z) dr
−
z (rn−1 Φ(z 0 ))0 −
ωn (q)
rn
rn−1 |z 0 (r)|p −
rn
r
= ωn (q) rn−1 Φ(z 0 )z |rn+1
n
= 0.
p
Consequently, by Corollary 1, equation (2) cannot have a positive solution in Ω R for
every R > 0.
The previous theorem shows that any oscillation criterion for the half-linear ordinary differential equation
0
(r(t)Φ(x0 )) + c(t)Φ(x) = 0
(16)
can be applied to (2). A typical example is the Leighton-Wintner criterion (see
e.g. [20] for its proof) which states that (16) is oscillatory provided
(17)
Z
∞
r
1−q
(t) dt = ∞ and
lim
t→∞
Z
t
c(s) ds = ∞.
587
The application of this criterion to (2) gives the following statement.
Theorem 4. Let p > n. If
lim
r→∞
Z
c(x) dx = ∞
kxkq 6r
then (2) possesses no positive solution in the exterior domain ΩR = {x ∈ R} for every R > 0.
n
: kxkq >
. By Theorem 3, it suffices to prove oscillation of (15), but it follows
immediately from the just mentioned Leighton-Wintner criterion (17) since (because
of p > n)
Z
Z
∞
and
lim
r→∞
Z
∞
r(n−1)(1−q) dr =
r
C(r) dr = lim
r→∞
Z
n−1
r− p−1 dr = ∞
c(x) dx = ∞.
kxkq 6r
The results of this section should be regarded only as “a sample” of the applications
of Picone’s identity in the oscillation theory of (2). For example, following [10], [14],
any oscillation and conjugacy criterion for the half-linear equation (15) (see., e.g.,
[8], [9]) yields a corresponding result on nonexistence of positive solutions of (2). In
addition, in our investigation we have till now completely neglected the relationship
between the existence of a nonzero solution of (2) and the solvability of (6). This
method of oscillation theory is usually regarded as the Riccati technique and was
used in the oscillation theory of PDE’s e.g. in [10], [18], [19], [21], [26] and other
papers. We hope to investigate applications of the Picone identity in the oscillation
theory of (2) as well as to look for a more general form of this identity in a subsequent
paper.
References
[1] W. Allegretto, Y. X. Huang: A Picone’s identity for the p-Laplacian and applications.
Nonlin. Anal. 32 (1998), 819–830.
[2] W. Allegretto, Y. X. Huang: Principal eigenvalues and Sturm comparison via Picone’s
identity. J. Differ. Equations 156 (1999), 427–438.
[3] M. Bohner: Linear Hamiltonian difference systems: disconjugacy and Jacobi-type conditions. J. Math. Anal. Appl. 199 (1996), 804–826.
[4] G. Bognár: A lower bound for smallest eigenvalue of some nonlinear eigenvalue problems
on convex domains in two dimensions. Appl. Math. 51 (1993), 277–288.
[5] G. Bognár: On the solution of some nonlinear boundary value problem. Publ. Univ.
Miskolc, Ser. D, Nat. Sci. Math. 36 (1995), 13–22.
588
[6] G. Bognár: Existence theorems for eigenvalues of a nonlinear eigenvalue problem. Commun. Appl. Nonlinear Anal. 4 (1997), 93–102.
[7] J. I. Díaz: Nonlinear Partial Differential Equations and Free Boundaries. Vol I. Elliptic
Equations, Pitman, London, 1985.
[8] O. Došlý: Oscillation criteria for half-linear second order differential equations. Hiroshima Math. J. 28 (1998), 507–521.
[9] O. Došlý: Methods of oscillation theory of half-linear second order differential equations.
Czechoslovak Math. J. 50 (2000), 657–671.
[10] O. Došlý, R. Mařík: Nonexistence of positive solutions of PDE’s with p-Laplacian. Acta
Math. Hungar. 90 (2001), 89–107.
[11] P. Drábek, A. Kufner, F. Nicolosi: Nonlinear Elliptic Equations—Singular and Degenerate Case. University of West Bohemia Press, Plzeň, 1996.
[12] D. R. Dunninger: A Sturm comparison theorem for some degenerate quasilinear eliptic
operators. Boll. UMI 9 (1995), 117–121.
[13] J. Jaroš, T. Kusano: A Picone type identity for second order half-linear differential equations. Acta Math. Univ. Comen. 68 (1999), 117–121.
[14] J. Jaroš, T. Kusano, N. Yosida: A Picone-type identity and Sturmian comparison and
oscillation theorems for a class of half-linear partial differential equations of the second
order. Nonlinear Anal. TMA 40 (2000), 381–395.
[15] K. Kreith: Oscillation Theory. Lectures Notes in Math. No. 324, Springer, Berlin, 1973.
[16] K. Kreith: Picone’s identity and generalizations. Rend. Mat. 8 (1975), 251–261.
[17] W. C. Kelley, A. C. Peterson: Difference Equations: An Introduction with Applications.
Acad. Press, San Diego, 1991.
[18] R. Mařík: Oscillation criteria for PDE with p-Laplacian via the Riccati technique. J.
Math. Anal. Appl. 248 (2000), 290–308.
[19] R. Mařík: Oscillation criteria for the Schrödinger PDE. Adv. Math. Sci. Appl. 10 (2000),
495–511.
[20] D. J. Mirzov: Asymptotic Properties of Solutions of Nonlinear Nonautonomous Ordinary
Differential Systems. Adygea, Maikop, 1993.
[21] W. F. Moss, J. Piepenbrick: Positive solutions of elliptic equations. Pacific J. Math. 75
(1978), 219–226.
[22] E. Müller-Pfeiffer: On the existence of nodal domain for elliptic differential operators.
Proc. Roy. Soc. Edinburgh 94A (1983), 287–299.
[23] E. Müller-Pfeiffer: An extension of the Sturm-Picone theorem to elliptic differential
equations. Proc. Roy. Soc. Edinburgh 97A (1984), 209–215.
[24] M. Picone: Sui valori eccezionalle di un parametro da cui dipende un’ equatione differenzialle lineare del secondo ordine. Ann. Scuola Norm. Sup. Pisa 11 (1910), 1–141.
[25] R. T. Rockafeller: Convex Analysis. Princeton, 1970.
[26] U. W. Schminke: The lower spectrum of Schrödinger operators. Arch. Rat. Mech. Anal.
75 (1981), 147–155.
[27] C. A. Swanson: Picone’s identity. Rend Mat. 8 (1975), 373–397.
Author’s address: Ondřej Došlý, Mathematical Institute, AS CR, Žižkova 22, CZ-616 62
Brno, e-mail: dosly@math.muni.cz.
589
Download